Font Size: a A A

Research On Iterative Learning Control Strategy Based On Partical Least Squares

Posted on:2014-11-22Degree:MasterType:Thesis
Country:ChinaCandidate:W W GuoFull Text:PDF
GTID:2268330425991817Subject:Control engineering
Abstract/Summary:PDF Full Text Request
It has great significance for product quality control to build an accurate model and combined control algorithms in industrial process. Partial Least Squares (PLS), as one of data-driven methods, has the unique advantage in modeling. Iterative Learning Control (ILC) is very applicable for batch process with repetitive nature. The two algorithms have been studied widely since they were proposed. However, researches on the two algorithms are relatively independent. Recently, researchers start to study the ILC combined with PLS, it has great significance for solving product quality control problem as the advantages of PLS modeling is used into the ILC effectively.This paper first introduces the theory of PLS and ILC, the main research contents based on them are as follows:1. This paper studies the traditional PLS algorithm in view of space decomposition in order to improve the accuracy of PLS modeling method in complex industrial process quality control, the analysis results show that the residual space of the input variable after PLS decomposition still contains related information to quality variables, the related information can affect the PLS modeling accuracy. This paper deals with the residual space, founds the relationship between the residual space and the output variable from the projection method. Then the paper propose the PLS algorithm with residual corrected. Compared with the traditional PLS algorithm, the improved PLS algorithm gives a more accurate relationship between input variables and output variables, improves the accuracy and predictive ability of the PLS model.2. To solve the modeling problem in batch processes, firstly, this paper expands the PLS with residual corrected to improved Multi-way PLS (MPLS), gets the accurate information of the object. Considering the mismatch between the model and the actual object, the model parameters uncertainty set is established, and the worst-case performance index is introduced, linear matrix inequality (LMI) is used to solve the worst-case performance index. Then the worst-case performance index MPLS-ILC is proposed. Because the improved MPLS can get the information of the object before control, and the uncertainty is considered, the proposed strategy can reject the deviations from the model mismatch and the disturbances that repeat themselves in the subsequent batches. Theoretical analysis and the simulation results of fused magnesium furnace batch process have shown the stability and robustness of the proposed method.
Keywords/Search Tags:iterative learning control, partial least squares, worst-case performance, X-residual space
PDF Full Text Request
Related items